brainstorm-mcp
brainstorm-mcp runs debates among GPT, Gemini, DeepSeek, Claude and Ollama models in your editor and returns a three-point verdict
- 73 ★ on the repository
- Last commit
- MIT
Extended access
Why this level
- In API mode code and context go to the chosen model providers, and keys live in configuration or the environment
- In CLI mode the server launches agent CLIs from PATH and uses up the limits of your subscriptions
Install
- Works in
In your terminal, with SkillFoxx CLI
npx skillfoxx add mcp/brainstorm-mcpDetects the agents on your machine, shows the access level and pins the version.
Other ways to install
Access and checks
How we reviewExtended accessThe tool runs code or makes network calls and only reads external data.
- In API mode code and context go to the chosen model providers, and keys live in configuration or the environment
- In CLI mode the server launches agent CLIs from PATH and uses up the limits of your subscriptions
This is third-party code. Review the repository files before installing.
What it does
The MCP server gathers several models and asks them to argue: each round they see one another's answers and critique them. The result is a three-point verdict: a recommendation, key tradeoffs and the strongest disagreement. The author's idea: one answer from one model hides what it is unsure about, whereas disagreement between three models is itself a signal. There are three modes. In hosted mode no keys are needed: the debate runs on models in your environment, for example Claude Opus, Sonnet and Haiku through subagents. In API mode the server gets keys and calls OpenAI, Gemini, DeepSeek, Groq, Mistral, Together, Moonshot, MiniMax, GLM, Qwen, xAI and local Ollama in parallel. In CLI mode the server launches agent CLIs from PATH such as `claude` and `codex` and uses your subscription instead of API credits. There are several tools: `brainstorm` for a full debate, `brainstorm_quick` for instant opinions in under 10 seconds and `brainstorm_review` for multi-model code review with finding severity and a verdict. Debate styles are freeform, red-team and Socratic. Code, diffs and architecture docs can be injected into the discussion. One model failing does not abort a debate, and Claude can take part as one of the debaters.
Who it is for. Developers and architects who need to compare several models' opinions before a decision, such as choosing between GraphQL and REST, instead of trusting one answer.
Good fit when
- You are choosing between GraphQL and REST and want three models critiquing each other with a three-point verdict
- Before merge you need a code review by several models with severity ratings and a shared verdict
- You want a red team: models hunt weak spots in a spec and attack the decision
- You need instant opinions from several models in 10 seconds on an architecture question
Not a fit when
- You need one quick answer: a multi-model debate uses noticeably more tokens and time
- Code must not go to third-party APIs: in API mode the text goes to model providers
- You have access to no model except Claude in your environment: only hosted mode remains, with no variety of opinions
Example request
Run a debate between opus, sonnet and haiku on whether to use GraphQL or REST for our API and give me the three-point verdictMCP
- Transport
- stdio
- Authentication
- key is optional
Environment variables
OPENAI_API_KEYsecret- OpenAI key for API mode
GEMINI_API_KEYsecret- Gemini key for API mode
DEEPSEEK_API_KEYsecret- DeepSeek key for API mode
BRAINSTORM_CONFIG- Path to a JSON file with providers and models
Limitations
- In API mode provider keys come from environment variables or a `BRAINSTORM_CONFIG` file, and code and context go to the chosen providers.
- CLI mode launches agent CLIs from your machine and spends their subscriptions, so watch the limits of those subscriptions.
- Multi-round debates consume many tokens: more models and rounds mean a bigger bill.
- Model names and versions in the README change quickly, so check current identifiers in provider documentation.
How to disable. Remove the server with claude mcp remove brainstorm or delete the brainstorm entry from `.mcp.json` and `claude_desktop_config.json`. A global install is removed with npm uninstall -g brainstorm-mcp. Provider keys live in your configuration or environment, so delete them there yourself.
FAQ
How do I connect brainstorm-mcp to Claude Code?
Run `claude mcp add brainstorm -- npx -y brainstorm-mcp` and the server works in hosted mode with no keys. For GPT, Gemini and DeepSeek add keys to the `env` block of `.mcp.json`.
Do I need API keys?
No, in hosted mode the debate runs on models in your environment without keys. Keys are for API mode: `OPENAI_API_KEY`, `GEMINI_API_KEY`, `DEEPSEEK_API_KEY` and others.
What is the three-point verdict?
A recommendation, key tradeoffs and the strongest disagreement among the models. It shows the decision and where the models disagreed.
How do I get a code review from several models?
Call the `brainstorm_review` tool: it sends code to several models and returns structured findings with severity and a verdict. Context can be code, a diff or a document.
What is CLI mode?
The server launches agent CLIs from PATH such as `claude` and `codex` and runs the debate through them. Spend goes against your subscription and no API credits are used, and detected CLIs register automatically at startup.
Discussion
No questions or reviews yet.
Related
MCP serversAn MCP server and terminal where several models debate a question: Claude, GPT, Gemini, Grok and Ollama use one of seven methods
Plugin and MCP server that asks GPT, Gemini, Grok and local models to review a plan or hunt for bugs before code ships.
An MCP server that sends one prompt to several OpenAI, Anthropic, Gemini, Groq, DeepSeek and Ollama models at once and can draw a decision from them
A local web service and MCP server that gathers several models to answer, peer review each other or debate as advisor personas.
An MCP server that lets Claude Code, Codex and other agents call Gemini, GPT, Grok and local models for review, debugging and consensus
A Claude Code plugin that asks several AI models the same question and shows their answers side by side with a synthesis of agreement and divergence
Comparison with similar tools
Cards that share the main search query, and how they differ.
| Tool | Official | Authentication | Repository stars | Main difference |
|---|---|---|---|---|
| brainstorm-mcp (this card) | no | key is optional | 73 | |
| LLM Council | no | 3,015 | a model council as a skill with no dedicated MCP server or code review mode | |
| Council of High Intelligence | no | 4,837 | role-based advisors on one model with no debates among different models |
README in short
The README starts with the idea: one model answer hides uncertainty while a debate among several exposes it. The quick start adds the server with one command and runs in hosted mode. Features are listed: three modes, a quick mode, code review, debate styles, context injection and the three-point verdict. Install for Claude Code, Claude Desktop and manual use follows, then provider setup through environment variables or a file, a model table and CLI mode on subscriptions.